Hi, sorry about the difficulty with Gemini 2.5 Pro here.
Just want to clear a couple things up:
1. Model Error w/ Resume button
In this case, if the model has errored and we offer you a resume button, you will not be charged for this request unless you choose to resume, which should let you continue where you left off. This situation is caused when, for whatever reason, there is an error in the generation of the response, but we can detect it and re-trigger it for you, without charging you double.
2. Model “stops” cleanly without any visible error
This is an Gemini-specific problem that we have pushed Google to fix for a few weeks now, but they seem to still be having issues. This happens when Google tells us (via the response their API sends) that the model has finished generating, but does not tell us that any error occurred during the process, even though it’s clear when reading the response that the model stopped midway.
This is unfortunately almost impossible for us to detect, as Google tells us it was a successful generation, and the only way to know if by a human reading the conversation.
We believe a fix should be available very soon (this week or next I would guess), but if you are seeing this regularly, I’d recommend switching to another model.
You will know when this specific issue occurs, because the usual icons at the end of the message will show with no errors when specifically using Gemini 2.5 Pro, as shown in @KamilTheDev’s screenshot.
